\r\n File \"/Users/hy/PycharmProjects/RAG-Anything/venv/lib/python3.11/site-packages/mineru/utils/models_download_utils.py\", line 21, in auto_download_and_get_model_root_path\r\n root_path = local_models_config.get(repo_mode, None)\r\n │ └ 'vlm'\r\n └ None\r\n\r\nAttributeError: 'NoneType' object has no attribute 'get'\r\n",
"created_at": "2026-01-23T10:08:44Z",
"updated_at": "2026-01-23T10:09:09Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "HYSMN",
"avatar_url": "https://avatars.githubusercontent.com/u/32283625?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AgaBp",
"number": 13,
"title": "Integration with existing LightRAG server that uses Ollama",
"body": "I have an existing LightRAG with a lot of content. I have been looking at whether something like Docling could improve the embedding as many of my PDFs are complicated and messy brochures...but RAGAnything seems to provide quite a nice preprocessing pipeline that one assumes will do what I'm looking for anyway.\n\nRAGAnything talks about use of existing LightRAG but I'm not clear how that works with LightRAG server and also as per another discussion quite how RAGAnything supports Ollama rather than OpenAI.\n\nI don't want to start again with a new LightRAG store as I'm not sure I can find all the documents I have uploaded nor do I really want to reprocess them just to see how RAGAnything improves ingest\n\nCan anyone help?",
"created_at": "2025-06-25T20:45:45Z",
"updated_at": "2025-12-23T21:42:28Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "helicalchris",
"avatar_url": "https://avatars.githubusercontent.com/u/35872772?u=f8246e7b084741f5dd66a1995fdab016bfb297bd&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AjYPe",
"number": 179,
"title": "加载已有的RAG库时报错。",
"body": "# Same working_dir as processing step!\r\n config = RAGAnythingConfig(working_dir=\"./rag_storage\")\r\n\r\n rag = RAGAnything(config=config, llm_model_func=llm_model_func, embedding_func=embedding_func)\r\n\r\n # Query existing processed data\r\n result = await rag.aquery(\r\n \"RAG-anything是什么\",\r\n mode=\"hybrid\"\r\n )\r\n print(result) \r\n运行出错: No LightRAG instance available. Please process documents first or provide a pre-initialized LightRAG instance.",
"created_at": "2025-12-23T01:57:52Z",
"updated_at": "2025-12-23T01:57:52Z",
"category": {
"name": "General",
"emoji": ":speech_balloon:"
},
"answer": null,
"user": {
"login": "Zhaofan2021",
"avatar_url": "https://avatars.githubusercontent.com/u/85424946?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AiOhm",
"number": 132,
"title": "Any better way to use RAG-Anything at Enterprises?",
"body": "Any better way to use RAG-Anything at Enterprises?\r\ni.e in AWS or Azure with Azure OpenAI or Claude",
"created_at": "2025-10-02T06:07:10Z",
"updated_at": "2025-12-22T10:14:16Z",
"category": {
"name": "General",
"emoji": ":speech_balloon:"
},
"answer": null,
"user": {
"login": "mahanteshimath",
"avatar_url": "https://avatars.githubusercontent.com/u/13198314?u=e92f6c8c54fe81ccb29a95578eb01e85b54dda2b&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4Ag8nQ",
"number": 66,
"title": "How RAGAnything query work",
"body": "\r\n![\"git\"](\"https://github.com/user-attachments/assets/bbefd75d-8d8a-439e-bc7a-728ed0e03542\")
\r\n\r\n\r\nI want to know — does the output_dir play any role in answering the query?\r\n\r\nI processed the document once and was able to query it successfully. But when I commented out the process() function and kept only the query() function, it gave an error. Why doesn’t it pick up the already processed document? The path is already defined in the raganything config.\r\n\r\nAlso, is there a way to query previously processed documents?\r\nWith LightRAG, I just provided the path and it worked.\r\nSo then my question is: what kind of data resides in the output_dir, and Does is it role in querying?\r\n\r\nwhy i am asking the role of output_dir is because that on first time process document function and query function is in the same file.\r\n",
"created_at": "2025-07-27T06:13:09Z",
"updated_at": "2025-12-15T16:59:33Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "vipan786",
"avatar_url": "https://avatars.githubusercontent.com/u/44468694?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AjJwg",
"number": 174,
"title": "Struggling to query table (check pdf at the bottom of the post)",
"body": "Hey guys,\r\n\r\nI am trying to query this table for about 3 weeks now. However, it was in vain. I tried a lot of configurations but i wasn't able to get 100% correct answers. Some of the rows turn out to be fully correct but others are not. It's because either 1) the LLM used the columns from another row or 2) mixed up values from different rows. Early on the LLM was even hallucinating values which are not part of the table. I managed to restrict that behaviour by:\r\n\r\n1. **Ultra-strict prompt**\r\n\r\n```\r\nmerged_prompt = f\"\"\"\r\nYou are a document-grounded assistant working on technical PDFs for mechanical components\r\n(e.g. spur gears, shafts, bearings).\r\n\r\nYou are given CONTEXT extracted from the PDF. CONTEXT may include:\r\n- OCR'd tables with dimensions and properties\r\n- Text paragraphs (materials, fits, tolerances, notes)\r\n- Images of 2D technical drawings portraying geometry\r\n- Descriptions of formulas and equations\r\n\r\nGLOBAL RULES (MUST OBEY)\r\n- Use ONLY information that appears in the CONTEXT.\r\n- You are NOT allowed to invent, modify, or \"correct\" any numeric value.\r\n- Do NOT use external standards, formulas, or domain knowledge.\r\n- Do NOT interpolate, smooth, or approximate values.\r\n- If you output a numeric value, it MUST literally appear somewhere in the CONTEXT\r\n(same digits, commas/dots, units, spacing).\r\n- If the information needed to answer the question is missing or incomplete, answer EXACTLY:\r\nCANNOT_FIND_ANSWER_IN_DOCUMENT\r\nand do not guess.\r\n\r\nSPECIAL CASE: STRICT TABLE ROW COPY (MUST FOLLOW EXACTLY)\r\n- If the user asks for a specific table row to be \"copied\", \"returned\", or \"extracted\"\r\n(e.g. “full row for Z2=25 teeth”, “give me the row where Z2 = 25”, etc.):\r\n1) Locate the `table_body` string inside CONTEXT (the HTML fragment starting with ).\r\n2) Inside this `table_body` HTML, search ONLY within the ...
rows.\r\n3) Find all ...
rows where the FIRST cell exactly matches the requested key\r\n (e.g. 25 for “Z2=25”; do NOT approximate, do NOT use closest value).\r\n4) If you find one or more matching rows:\r\n - OUTPUT ONLY the exact characters of each matching \" | ...
\" substring from the input.\r\n - Preserve all characters exactly: tags, spaces, commas, dots, decimal separators, units, and line breaks.\r\n - Do NOT wrap the output in Markdown, tables, code fences, or any extra text.\r\n - Do NOT reorder columns. Do NOT explain the result. Just output the raw ...
line(s).\r\n5) If you do NOT find any matching row:\r\n - Output EXACTLY: NOT_IN_CONTEXT\r\n - Do NOT construct a plausible row, do NOT guess, and do NOT reuse another row as a template.\r\n\r\nGENERAL TASK (NON-STRICT CASES)\r\n- For all other questions:\r\n- Read the user query.\r\n- Search the CONTEXT (tables, drawings, text, equations) for information relevant to the query.\r\n- You may combine information from tables, drawings, equations, and text,\r\n but NEVER create new numerical values.\r\n- Prefer precise table entries and dimension annotations from drawings over vague text.\r\n\r\nOUTPUT FORMAT FOR GENERAL QUESTIONS\r\n- When listing dimensions, use bullets:\r\n- = — \r\n\r\nCONTEXT START\r\n{system_prompt}\r\nCONTEXT END\r\n\r\nUser query:\r\n{prompt or \"\"}\r\n\"\"\"\r\n\r\nsystem_prompt = (\r\n \"You are a deterministic, document-grounded assistant. \"\r\n \"You must treat the CONTEXT as the only source of truth and you must never invent, \"\r\n \"modify, or reformat numeric values. When the STRICT TABLE ROW COPY rules apply, \"\r\n \"you MUST follow them exactly.\"\r\n)\r\n```\r\n2. **Use this configuration:**\r\n```\r\nconfig = RAGAnythingConfig(\r\n working_dir=\"./rag_storage_multimodal_ollama\",\r\n parser=\"mineru\",\r\n parse_method=\"auto\",\r\n enable_image_processing=True,\r\n enable_table_processing=True,\r\n enable_equation_processing=True,\r\n context_mode=\"chunk\",\r\n max_context_tokens=4096, # enough for the full table chunk\r\n context_window=0, # just that page\r\n include_headers = True, # Include headers for technical content\r\n include_captions = True, # Include captions for images/tables\r\n context_filter_content_types = [\"text\", \"table\"] \r\n)\r\nrag = RAGAnything(\r\n config=config,\r\n lightrag_kwargs = { \"enable_llm_cache\": False,\r\n \"llm_model_kwargs\": \r\n {\"options\": {\r\n \"num_ctx\": 32768,\r\n \"temperature\": 0.05, # almost deterministic\r\n \"top_p\": 0.8, # reduce tail sampling\r\n \"top_k\": 40, # limit to top 40 probable tokens\r\n \"repeat_penalty\": 1.05 # mild, to avoid weird loops\r\n }\r\n }\r\n },\r\n llm_model_func=ollama_llm_model,\r\n vision_model_func=vision_model_func,\r\n embedding_func=EmbeddingFunc(\r\n embedding_dim=768,\r\n func=lambda texts: ollama_embed(texts, embed_model=EMBEDDING_MODEL_NAME)\r\n )\r\n)\r\ntext_result = await rag.aquery(\r\n \"From the table with dimensions for polyacetal spur gears, please extract the full row for ZZ=60 teeth\",\r\n vlm_enhanced=False,\r\n enable_rerank=False,\r\n chunk_top_k=6, # because my pdf contains 2 imgs, 1 table, 1 txt paragraph and 2 discarded elements\r\n mode=\"naive\",\r\n )\r\n```\r\n---\r\n\r\nWhat am I doing wrong? I read that all current LLMs are not reliably 100% when given a long table in the context and ask it “give me the entire row where Z2=12”. any workaround?\r\n\r\nP.S: \r\n> 1) The model sees all context at query time\r\n> 2) I am using ollama local model llama3.2-vision:11b-instruct-q4_K_M for both \"llm_model_func\" and \"vision_model_func\" and nomic-embed-text\" for embeddings.\r\n[gear_span.pdf](https://github.com/user-attachments/files/23988497/gear_span.pdf)\r\n",
"created_at": "2025-12-06T14:58:15Z",
"updated_at": "2025-12-06T14:58:15Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "Raouf71",
"avatar_url": "https://avatars.githubusercontent.com/u/72918044?u=29585180ae7c880806949d3f4f9a7d352963567f&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AhmlN",
"number": 96,
"title": "Does RAG Anything supports multiple documents?",
"body": "Hi!\r\n\r\nDoes this library support multiple documents?\r\n\r\nAs per below code, it only supports single document\r\n\r\n # Process a document\r\n await rag.process_document_complete(\r\n file_path=\"path/to/your/document.pdf\",\r\n output_dir=\"./output\",\r\n parse_method=\"auto\"\r\n )\r\n\r\n\r\nThank you\r\n",
"created_at": "2025-08-28T01:11:21Z",
"updated_at": "2025-11-28T04:58:22Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "naveenoxygen",
"avatar_url": "https://avatars.githubusercontent.com/u/165016822?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AgTz_",
"number": 5,
"title": "Does this have ollama support?",
"body": "Does this have ollama support?",
"created_at": "2025-06-18T19:21:53Z",
"updated_at": "2025-11-27T20:37:11Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "nasaisrllycool",
"avatar_url": "https://avatars.githubusercontent.com/u/215014816?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AjB2L",
"number": 169,
"title": "Scalable RAG",
"body": "Are the project planning to extend the document ingestion RAG-Anything to be more scalable? For example with Ray or Anyscale.",
"created_at": "2025-11-26T20:18:14Z",
"updated_at": "2025-11-26T20:18:15Z",
"category": {
"name": "Ideas",
"emoji": ":bulb:"
},
"answer": null,
"user": {
"login": "r3006",
"avatar_url": "https://avatars.githubusercontent.com/u/49816401?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AizZG",
"number": 160,
"title": "Does RAG-anything support querying between different rag_storage",
"body": "I mean, if I build two rag_storage paths with different files, how could I query by path indicated?\r\n```python\r\nlightrag_instance = LightRAG(working_dir='/user/namespace_1',\r\n llm_model_func=llm_model_func,\r\n embedding_func=embedding_func)\r\n```\r\nIn this example, LightRAG only receives a single path\r\n",
"created_at": "2025-11-10T04:41:42Z",
"updated_at": "2025-11-10T04:41:42Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "Chen1303005809",
"avatar_url": "https://avatars.githubusercontent.com/u/59914914?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AiraM",
"number": 152,
"title": "Support for Chunk Metadata when dealing with text",
"body": "Hey everyone —\r\nI’ve been using RAG-Anything, and while it’s great for multimodal stuff (text, tables, images, etc.), I’m hitting a wall with chunk metadata.\r\n\r\nYou can extract metadata for non-text elements, but there doesn’t seem to be any way to know which doc or page a text chunk actually came from. Once it’s chunked, all that context seems gone.\r\n\r\nAm I missing something? Or is this just not supported right now?",
"created_at": "2025-10-30T23:31:43Z",
"updated_at": "2025-10-30T23:31:43Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "DLOVRIC2",
"avatar_url": "https://avatars.githubusercontent.com/u/66421606?u=20b7137c45b4589868ae3d45cb84cdfc67268074&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AinSz",
"number": 148,
"title": "Support for Supabase Vector Database",
"body": "Is there plan to integrate this framework to work with Supabase Vector Stores?",
"created_at": "2025-10-26T19:23:01Z",
"updated_at": "2025-10-26T19:23:02Z",
"category": {
"name": "General",
"emoji": ":speech_balloon:"
},
"answer": null,
"user": {
"login": "JDBem",
"avatar_url": "https://avatars.githubusercontent.com/u/74302008?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AidDs",
"number": 141,
"title": "How to fix this on gg colab",
"body": "I followed the path to install and run the first usage explanation to try it out. I replaced the api key and document path but there is more problems\r\nINFO: RAGAnything initialized with config:\r\nINFO: Working directory: ./rag_storage\r\nINFO: Parser: mineru\r\nINFO: Parse method: auto\r\nINFO: Multimodal processing - Image: True, Table: True, Equation: True\r\nINFO: Max concurrent files: 1\r\nINFO: Parser 'mineru' installation verified\r\nINFO: Initializing LightRAG with parameters: {'working_dir': './rag_storage'}\r\nINFO: [_] Created new empty graph file: ./rag_storage/graph_chunk_entity_relation.graphml\r\nINFO: Multimodal processors initialized with context support\r\nINFO: Available processors: ['image', 'table', 'equation', 'generic']\r\nINFO: Context configuration: ContextConfig(context_window=1, context_mode='page', max_context_tokens=2000, include_headers=True, include_captions=True, filter_content_types=['text'])\r\nINFO: LightRAG, parse cache, and multimodal processors initialized\r\nINFO: Starting complete document processing: /1307045.pdf\r\nINFO: Starting document parsing: /1307045.pdf\r\nINFO: Using cached parsing result for: /1307045.pdf\r\nINFO: * Total blocks in cached content_list: 56\r\nINFO: Content separation complete:\r\nINFO: - Text content length: 23326 characters\r\nINFO: - Multimodal items count: 4\r\nINFO: - Multimodal type distribution: {'image': 3, 'table': 1}\r\nINFO: Setting content source for context-aware multimodal processing...\r\nINFO: Content source set with format: minerU\r\nINFO: Content source set with format: minerU\r\nINFO: Content source set with format: minerU\r\nINFO: Content source set with format: minerU\r\nINFO: Content source set for context extraction (format: minerU)\r\nINFO: Starting text content insertion into LightRAG...\r\nWARNING: Ignoring document ID (already exists): doc-50d7ecd2cdb2e73645fbbdd75c97184d (1307045.pdf)\r\nWARNING: No new unique documents were found.\r\n---------------------------------------------------------------------------\r\nTypeError Traceback (most recent call last)\r\n[/tmp/ipython-input-4020778013.py](https://localhost:8080/#) in ()\r\n 128 print(\"Multimodal query result:\", multimodal_result)\r\n 129 \r\n--> 130 await main()\r\n\r\n5 frames\r\n[/tmp/ipython-input-4020778013.py](https://localhost:8080/#) in main()\r\n 102 \r\n 103 # Process a document\r\n--> 104 await rag.process_document_complete(\r\n 105 file_path=\"/1307045.pdf\",\r\n 106 output_dir=\"./output\",\r\n\r\n[/content/RAG-Anything/raganything/processor.py](https://localhost:8080/#) in process_document_complete(self, file_path, output_dir, parse_method, display_stats, split_by_character, split_by_character_only, doc_id, **kwargs)\r\n 1475 if text_content.strip():\r\n 1476 file_name = os.path.basename(file_path)\r\n-> 1477 await insert_text_content(\r\n 1478 self.lightrag,\r\n 1479 input=text_content,\r\n\r\n[/content/RAG-Anything/raganything/utils.py](https://localhost:8080/#) in insert_text_content(lightrag, input, split_by_character, split_by_character_only, ids, file_paths)\r\n 164 \r\n 165 # Use LightRAG's insert method with all parameters\r\n--> 166 await lightrag.ainsert(\r\n 167 input=input,\r\n 168 file_paths=file_paths,\r\n\r\n[/usr/local/lib/python3.12/dist-packages/lightrag/lightrag.py](https://localhost:8080/#) in ainsert(self, input, split_by_character, split_by_character_only, ids, file_paths, track_id)\r\n 928 \r\n 929 await self.apipeline_enqueue_documents(input, ids, file_paths, track_id)\r\n--> 930 await self.apipeline_process_enqueue_documents(\r\n 931 split_by_character, split_by_character_only\r\n 932 )\r\n\r\n[/usr/local/lib/python3.12/dist-packages/lightrag/lightrag.py](https://localhost:8080/#) in apipeline_process_enqueue_documents(self, split_by_character, split_by_character_only)\r\n 1379 # Ensure only one worker is processing documents\r\n 1380 if not pipeline_status.get(\"busy\", False):\r\n-> 1381 processing_docs, failed_docs, pending_docs = await asyncio.gather(\r\n 1382 self.doc_status.get_docs_by_status(DocStatus.PROCESSING),\r\n 1383 self.doc_status.get_docs_by_status(DocStatus.FAILED),\r\n\r\n[/usr/local/lib/python3.12/dist-packages/lightrag/kg/json_doc_status_impl.py](https://localhost:8080/#) in get_docs_by_status(self, status)\r\n 116 if \"error_msg\" not in data:\r\n 117 data[\"error_msg\"] = None\r\n--> 118 result[k] = DocProcessingStatus(**data)\r\n 119 except KeyError as e:\r\n 120 logger.error(\r\n\r\nTypeError: DocProcessingStatus.__init__() got an unexpected keyword argument 'multimodal_processed' ",
"created_at": "2025-10-16T02:25:49Z",
"updated_at": "2025-10-16T02:25:50Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "alexdan1816",
"avatar_url": "https://avatars.githubusercontent.com/u/170811406?u=e2b5bd6be4e459013904dd16e49b37f966c2f81f&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AiaFe",
"number": 137,
"title": "How can `Rag Anyhting` connect with `Milvus data base` ?",
"body": " Load .env first\r\nload_dotenv()\r\nAPI_KEY = os.environ.get(\"OPENAI_API_KEY\")\r\n\r\nasync def build_lightrag():\r\n light = LightRAG(\r\n working_dir=\"./rag_storage\",\r\n vector_storage=\"MilvusVectorDBStorage\",\r\n workspace=os.getenv(\"MILVUS_WORKSPACE\"),\r\n llm_model_func=lambda prompt, system_prompt=None,history_messages=[],**kwargs: openai_complete_if_cache(\r\n \"gpt-4o\",\r\n prompt,\r\n system_prompt=system_prompt,\r\n history_messages=history_messages,\r\n api_key=API_KEY,\r\n ),\r\n embedding_func=EmbeddingFunc(\r\n embedding_dim=3072,\r\n max_token_size=8192,\r\n func=lambda texts: openai_embed(\r\n texts,\r\n model=\"text-embedding-3-large\",\r\n api_key=API_KEY,\r\n )\r\n ),\r\n )\r\n # Required LightRAG initialization\r\n await light.lightrag.initialize_storages()\r\n await initialize_pipeline_status()\r\n return light\r\n\r\n\r\ndef build_rag() -> RAGAnything:\r\n lightrag_instance = build_lightrag()\r\n config = RAGAnythingConfig(\r\n working_dir=\"./rag_storage\",\r\n parser=\"generic\",\r\n parse_method=\"auto\",\r\n enable_image_processing=False,\r\n enable_table_processing=False,\r\n enable_equation_processing=False,\r\n )\r\n return RAGAnything(lightrag=lightrag_instance, config=config) # RAG-Anything on LightRAG \r\n\r\nasync def main():\r\n ap = argparse.ArgumentParser()\r\n ap.add_argument(\"--file\", default=\"./rag_storage\")\r\n ap.add_argument(\"--question\", default=\"what is a vector database in this milvus?\")\r\n args = ap.parse_args()\r\n\r\n rag = build_rag()\r\n\r\n # # Required LightRAG initialization\r\n # await rag.lightrag.initialize_storages()\r\n # await initialize_pipeline_status()\r\n\r\n # Ingest and query via Milvus-backed LightRAG \r\n await rag.process_document_complete(\r\n file_path=args.file,\r\n output_dir=\"./output\",\r\n parse_method=\"auto\",\r\n )\r\n result = await rag.aquery_with_multimodal(args.question, mode=\"hybrid\")\r\n print(result)\r\n\r\nif __name__ == \"__main__\":\r\n asyncio.run(main())int(result)\r\n\r\nif __name__ == \"__main__\":\r\n asyncio.run(main())",
"created_at": "2025-10-13T05:34:07Z",
"updated_at": "2025-10-13T05:43:48Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "AMRENDRASINGH-COM",
"avatar_url": "https://avatars.githubusercontent.com/u/172993640?u=6134116841acd0d1219383ccccfbeeae5a30ce49&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AiHJ1",
"number": 123,
"title": "LightRAG Instance",
"body": "I'm using RAG-Anything and was able to embed a few documents. \r\n\r\nHow can I \"fire up\" the LightRAG interface to handle the data I have processed using RAG-Anything?",
"created_at": "2025-09-24T17:45:21Z",
"updated_at": "2025-09-28T17:35:35Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": {
"id": "DC_kwDOO3Bfkc4A3bN9",
"body": "🚀 **Great question!** RAG-Anything and LightRAG can work together seamlessly. Here's how to access your processed data through the LightRAG interface:\r\n\r\n## 🔧 **Method 1: Direct Instance Access**\r\n\r\nAfter processing documents with RAG-Anything, you can access the underlying LightRAG instance:\r\n\r\n```python\r\nfrom raganything import RAGAnything, RAGAnythingConfig\r\n\r\n# Initialize RAG-Anything\r\nconfig = RAGAnythingConfig(working_dir=\"./rag_storage\")\r\nrag = RAGAnything(config=config, llm_model_func=your_llm, embedding_func=your_embedding)\r\n\r\n# Process your documents\r\nawait rag.process_document_complete(\r\n file_path=\"document.pdf\",\r\n output_dir=\"./output\"\r\n)\r\n\r\n# Access the LightRAG instance directly\r\nlightrag_instance = rag.lightrag\r\n\r\n# Now you can use LightRAG methods directly\r\nresult = await lightrag_instance.aquery(\"Your query here\")\r\nprint(result)\r\n```\r\n\r\n## 📊 **Method 2: Using Existing Storage**\r\n\r\nIf you've already processed documents, you can initialize LightRAG pointing to the same storage:\r\n\r\n```python\r\nfrom lightrag import LightRAG, QueryParam\r\n\r\n# Point to the same working directory\r\nlightrag = LightRAG(\r\n working_dir=\"./rag_storage\", # Same as RAG-Anything working_dir\r\n llm_model_func=your_llm_function,\r\n embedding_func=your_embedding_function\r\n)\r\n\r\n# Query your processed data\r\nresult = await lightrag.aquery(\r\n \"What are the key insights from the processed documents?\",\r\n param=QueryParam(mode=\"hybrid\")\r\n)\r\n```\r\n\r\n## 🔍 **Method 3: Graph Visualization Interface**\r\n\r\nFor visual exploration of your knowledge graph:\r\n\r\n```python\r\n# After processing with RAG-Anything\r\nfrom lightrag.utils import EmbeddingFunc\r\nimport networkx as nx\r\n\r\n# Access the graph data\r\ngraph_data = rag.lightrag.chunk_entity_relation_graph\r\n\r\n# Create visualization\r\ndef visualize_knowledge_graph():\r\n G = nx.Graph()\r\n \r\n # Add nodes and edges from your processed data\r\n for entity in graph_data.entities:\r\n G.add_node(entity.name, **entity.properties)\r\n \r\n for relation in graph_data.relations:\r\n G.add_edge(relation.source, relation.target, \r\n weight=relation.strength)\r\n \r\n return G\r\n\r\n# Generate and display graph\r\ngraph = visualize_knowledge_graph()\r\n```\r\n\r\n## ⚙️ **Method 4: Custom Query Interface**\r\n\r\nBuild a simple interface to interact with your processed data:\r\n\r\n```python\r\nimport asyncio\r\n\r\nclass RAGInterface:\r\n def __init__(self, rag_anything_instance):\r\n self.rag = rag_anything_instance\r\n self.lightrag = rag_anything_instance.lightrag\r\n \r\n async def interactive_query(self):\r\n print(\"🤖 RAG-Anything Interactive Interface\")\r\n print(\"Type 'exit' to quit\")\r\n \r\n while True:\r\n query = input(\"\\n📝 Enter your query: \")\r\n if query.lower() == 'exit':\r\n break\r\n \r\n try:\r\n # Use different query modes\r\n hybrid_result = await self.lightrag.aquery(query, mode=\"hybrid\")\r\n print(f\"\\n🔍 Hybrid Search Result:\\n{hybrid_result}\")\r\n \r\n except Exception as e:\r\n print(f\"❌ Error: {e}\")\r\n\r\n# Usage\r\ninterface = RAGInterface(rag)\r\nawait interface.interactive_query()\r\n```\r\n\r\n## 🛠️ **Configuration Alignment**\r\n\r\nMake sure your LightRAG configuration matches RAG-Anything settings:\r\n\r\n```python\r\n# Check RAG-Anything config\r\nprint(f\"Working directory: {rag.config.working_dir}\")\r\nprint(f\"Enable image processing: {rag.config.enable_image_processing}\")\r\n\r\n# Configure LightRAG with same settings\r\nlightrag_config = {\r\n \"working_dir\": rag.config.working_dir,\r\n \"enable_image\": rag.config.enable_image_processing,\r\n \"enable_table\": rag.config.enable_table_processing,\r\n \"chunk_token_size\": rag.config.chunk_token_size,\r\n \"chunk_overlap_token_size\": rag.config.chunk_overlap_token_size\r\n}\r\n```\r\n\r\n## 🚨 **Troubleshooting Common Issues**\r\n\r\n1. **Storage Path Mismatch**:\r\n```python\r\n# Ensure paths match exactly\r\nassert rag.config.working_dir == lightrag.working_dir\r\n```\r\n\r\n2. **Missing Dependencies**:\r\n```bash\r\npip install lightrag networkx matplotlib plotly\r\n```\r\n\r\n3. **Memory Issues with Large Datasets**:\r\n```python\r\n# Use streaming queries for large datasets\r\nasync def stream_query(query):\r\n for chunk in await lightrag.aquery_stream(query):\r\n yield chunk\r\n```\r\n\r\n## 📈 **Advanced Integration Example**\r\n\r\nHere's a complete example combining both systems:\r\n\r\n```python\r\nimport asyncio\r\nfrom raganything import RAGAnything, RAGAnythingConfig\r\nfrom lightrag import LightRAG\r\n\r\nasync def full_rag_workflow():\r\n # Step 1: Process documents with RAG-Anything\r\n config = RAGAnythingConfig(\r\n working_dir=\"./unified_rag_storage\",\r\n enable_image_processing=True,\r\n enable_table_processing=True\r\n )\r\n \r\n rag_anything = RAGAnything(config=config)\r\n \r\n # Process your documents\r\n await rag_anything.process_folder_complete(\"./documents\")\r\n \r\n # Step 2: Access via LightRAG interface\r\n lightrag = rag_anything.lightrag\r\n \r\n # Step 3: Advanced querying\r\n queries = [\r\n \"Summarize the main topics across all documents\",\r\n \"What are the key relationships between entities?\",\r\n \"Find contradictions or inconsistencies in the data\"\r\n ]\r\n \r\n results = {}\r\n for query in queries:\r\n result = await lightrag.aquery(\r\n query, \r\n mode=\"hybrid\",\r\n only_need_context=False\r\n )\r\n results[query] = result\r\n \r\n return results\r\n\r\n# Run the workflow\r\nresults = await full_rag_workflow()\r\nfor query, result in results.items():\r\n print(f\"Query: {query}\")\r\n print(f\"Result: {result}\\n{'-'*50}\\n\")\r\n```\r\n\r\n## 💡 **Pro Tips**\r\n\r\n- **Use `hybrid` mode** for best results when querying processed RAG-Anything data\r\n- **Check working_dir consistency** between RAG-Anything and LightRAG instances\r\n- **Monitor memory usage** with large document sets\r\n- **Use async/await** for better performance with multiple queries\r\n\r\nThe key is that RAG-Anything builds on top of LightRAG, so you can access the underlying LightRAG instance directly or create a new one pointing to the same storage directory! 🎯\r\n\r\nLet me know if you need help with any specific integration scenario!"
},
"user": {
"login": "baartho",
"avatar_url": "https://avatars.githubusercontent.com/u/19720384?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AiF3i",
"number": 121,
"title": "ChromaDB Deprecated ?",
"body": "I want to use ChromaDB as the vector DB. Is it supported or deprecated ? ",
"created_at": "2025-09-23T12:33:34Z",
"updated_at": "2025-09-26T22:19:38Z",
"category": {
"name": "Q&A",
"emoji": ":pray:"
},
"answer": null,
"user": {
"login": "gstazure",
"avatar_url": "https://avatars.githubusercontent.com/u/27543824?v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AhQ9A",
"number": 81,
"title": "Rag anything working with LightRAG Server and WebUI",
"body": "I would like to ask if the raganything extension can work with the LightRAG Server and WebUI properly with the full capabilities of the raganything module. ",
"created_at": "2025-08-12T14:01:54Z",
"updated_at": "2025-09-26T15:16:30Z",
"category": {
"name": "General",
"emoji": ":speech_balloon:"
},
"answer": null,
"user": {
"login": "thanbskt",
"avatar_url": "https://avatars.githubusercontent.com/u/51063695?u=4a4b8bb79f03f9baf18840ea7c439aacc870f87b&v=4"
}
},
{
"id": "D_kwDOO3Bfkc4AgaNr",
"number": 16,
"title": "CUDA accelerate MinerU in process_folder_complete",
"body": "It is explained that device=cuda flag in process_document_complete can enable CUDA acceleration for MinerU, but this does not seem to work in the process_folder_complete function. What's the best way to approach this?",
"created_at": "2025-06-26T03:10:52Z",
"updated_at": "2025-06-26T03:10:53Z",
"category": {
"name": "General",
"emoji": ":speech_balloon:"
},
"answer": null,
"user": {
"login": "arjun-vdc",
"avatar_url": "https://avatars.githubusercontent.com/u/187897619?v=4"
}
}
],
"details": {
"id": 997220241,
"node_id": "R_kgDOO3BfkQ",
"name": "RAG-Anything",
"full_name": "HKUDS/RAG-Anything",
"private": false,
"owner": {
"login": "HKUDS",
"id": 118165258,
"node_id": "O_kgDOBwsPCg",
"avatar_url": "https://avatars.githubusercontent.com/u/118165258?v=4",
"gravatar_id": "",
"url": "https://api.github.com/users/HKUDS",
"type": "Organization",
"user_view_type": "public",
"site_admin": false
},
"description": "\"RAG-Anything: All-in-One RAG Framework\"",
"fork": false,
"url": "https://api.github.com/repos/HKUDS/RAG-Anything",
"created_at": "2025-06-06T06:47:29Z",
"updated_at": "2026-02-17T02:38:43Z",
"pushed_at": "2026-01-26T09:09:21Z",
"homepage": "http://arxiv.org/abs/2510.12323",
"size": 3273,
"stargazers_count": 13485,
"watchers_count": 13485,
"language": "Python",
"has_issues": true,
"has_projects": true,
"has_downloads": true,
"has_wiki": false,
"has_pages": false,
"has_discussions": true,
"forks_count": 1612,
"archived": false,
"disabled": false,
"open_issues_count": 102,
"license": {
"key": "mit",
"name": "MIT License",
"spdx_id": "MIT",
"url": "https://api.github.com/licenses/mit",
"node_id": "MDc6TGljZW5zZTEz"
},
"allow_forking": true,
"is_template": false,
"web_commit_signoff_required": false,
"has_pull_requests": true,
"pull_request_creation_policy": "all",
"topics": {
"0": "multi-modal-rag",
"1": "retrieval-augmented-generation"
},
"visibility": "public",
"forks": 1612,
"open_issues": 102,
"watchers": 13485,
"default_branch": "main",
"permissions": {
"admin": false,
"maintain": false,
"push": false,
"triage": false,
"pull": true
},
"temp_clone_token": "",
"custom_properties": {},
"organization": {
"login": "HKUDS",
"id": 118165258,
"node_id": "O_kgDOBwsPCg",
"avatar_url": "https://avatars.githubusercontent.com/u/118165258?v=4",
"gravatar_id": "",
"url": "https://api.github.com/users/HKUDS",
"type": "Organization",
"user_view_type": "public",
"site_admin": false
},
"network_count": 1612,
"subscribers_count": 86
},
"lastFetched": 1771297611507
} |